X hits on this document

PDF document

A Prototype Optical Tracking System Investigation and Development - page 128 / 170





128 / 170


Results and Discussion

deviations and assuming Gaussian statistics, this corresponds to an X resolution at 2 m of ±56.4 µm with 95% confidence. If the system is calibrated every 2 minutes then using the same 95% confidence interval this corresponds to a X resolution at 2 m of at best ±18.6 µm. Based on this, the system could operate without continuous calibration but it would be more favourable for the system to recalibrate often to obtain the best noise performance.


Comparison between modules

Module 2 and Module 4 were constructed using the MT9V023 and Module 3 was con- structed using the MT9V022, an older version of the sensor. Statistically it is not possible to draw conclusions about the performance between the two image sensors as the sample size is too small. However, the long term data shows that all three modules have similar performance. The best performance for the two minute interval data in the Y direction ranges between 0.0045 pixels for Module 4 and 0.0078 pixels for Module 3. s seen in Ta- ble 8.3, Module 3 usually has poorer performance than the other two modules but without a larger sample size it cannot be concluded with certainty that this is because of the older

image sensor.


Background noise

Background noise can have a large influence on the centroids.

Figure 8.11 was obtained

during testing. This measurement contains data from one marker but two groups of cen- troids can be seen. This can be explained by the position of the ROI that surrounds the marker. lthough the marker is stationary, the ROI moves throughout the measurement. The explanation for this is that noise on the marker image produces a variation in centroid positions causing the ROI to shift back and forth by one pixel. It has been determined that this shift in ROI position can cause multiple discrete centroid clouds if there is sufficient

background noise.

To explain this conclusion further, consider the ROI shown in Figure 8.12. Noise on the data causes the ROI to move back and forth between adjacent pixels. This ROI moves by 1 pixel left and right and this is the underlying reason that two groups of centroids can occur. The ROI can be split into three parts based on this movement. Part 2 is the portion of the region that stays constant. Part 1 is the portion of the ROI when it is in the left position and Part 3 is the portion when it is in the right position. The centroids of these three parts can be calculated as are shown in Figure 8.12 by the crosses. When the ROI is in the left position it consists of Part 1 and Part 2 and when it is in the right position it consists of Part 2 and Part 3. In the left position, the centroid is biased to the left by Part 1 and in the right position the centroid is biased to the right by Part 3. Therefore, if the ROI moves due to variation in the centroid position, then this movement can cause the two bunches

Document info
Document views507
Page views507
Page last viewedWed Jan 18 10:42:15 UTC 2017